ImgIng’s everyday tools run locally in your browser.
From compression and conversion to online design, AI cutout, animation creation and frame editing, ImgIng puts a complete workflow in the browser. Multi-layer layout, background setup and full-size export stay local; only a few formats the browser truly cannot handle use a server.
JPG, PNG, WebP, GIF and APNG use your device directly; AVIF writing and HEIC reading stay local after on-demand WASM loading.
Use PNG-8, local AI mask refinement and a multi-layer canvas with move, scale, background, undo and full-size local export.
Compress, create and edit GIF, APNG, animated WebP and animated AVIF; specialist formats have a clearly labelled server fallback.
Specialist server formats are paid features · admin@datadance.com
One image · Up to 40 MB
Instant = Browser can read the source and encode the target → processed locally, with no upload · Local WASM ↓ = WASM loads once (AVIF write / HEIC read), then processing remains local · Server = Only formats unavailable to the browser and local WASM (such as HEIC write) use a server
Confirm the suggested file name and location when saving
Same-format compression:Input and output formats stay the same · Up to 40 MB
Compress GIF / APNG directly · or select images to animate
Confirm the suggested file name and location when saving
This content has changed. Confirm that you meant to exit.
Formats are grouped by where they run: common formats stay in your browser, while specialist and legacy formats use a clearly labelled server path.
Outlined chips (HEIC read and AVIF write) download libheif / libavif only when first needed, then continue entirely on-device. Animated AVIF* uses the ImgIng WebCodecs AV1 delta-frame path with no alpha and infinite looping; unsupported browsers fall back to the server. Total input coverage: 7 + 9 = 16 formats; local outputs: 8 formats (4 still + 4 animated).
GIF, APNG, animated WebP and animated AVIF can all be produced on-device when supported. Compare quality, size and compatibility before choosing.
Still images, multilingual Image OCR, multi-layer design, animation and AI share one local-first engine, with processing boundaries stated before you start.
Palette refinement, multi-level alpha, optional dithering and quality controls. The 45.8 dB PNG-8 number is a single regression reference—not a universal promise.
Three verified PP-OCRv6 tiers run in a dedicated local Worker. Copy and TXT preserve source rows and separate detected table columns with tabs; Korean recognition is not claimed by the current models.
Use up to 24 editable image or text layers, load language fonts on demand, arrange every layer and export at full size without uploading assets.
Three local AI tiers preserve full alpha. Use a true 1 px brush, then set a colour or local background image, arrange the subject and add a multilingual text watermark without uploading pixels.
Preserve original timing and loops, reorder or delete frames, change duration and bundle GIF, APNG and WebP outputs.
ImgIng attempts actual decode and encode paths instead of inferring support from a browser name alone.
Zoom, pan and drag the fixed divider, then decode the real exported Blob to verify the file you will download.
No single product wins every job. This table states where ImgIng fits and where another tool may be stronger.
| Area | ImgIng | Squoosh | TinyPNG | Cloud converter |
|---|---|---|---|---|
| Primary processing | Device-first | Device | Server | Server |
| Still compression | PNG-8 / JPG / WebP / AVIF | Excellent codec controls | Mature cloud compression | Varies |
| Animation | Four formats + frame editor | Not a primary feature | Partial | Varies |
| HEIC / AVIF | Local HEIC read + AVIF write | Strong AVIF support | Not a primary feature | Broad, requires upload |
| AI background removal | Three local models + refinement | Not a primary feature | Not a primary feature | Varies |
| Account / watermark | None for device tools | None | Plan-dependent | Often limited |
Squoosh remains excellent for one-image codec experimentation. TinyPNG offers a mature cloud API ecosystem. Cloud converters cover more document and legacy formats. ImgIng focuses on local-first stills, animation, HEIC / AVIF and local AI in one workflow.
ImgIng tests whether your current device can really decode the input and encode the requested output. Unsupported paths are labelled before processing.
| Capability | Chrome / Edge | Firefox | Safari |
|---|---|---|---|
| PNG-8 / JPG / WebP | Supported | Supported | Version-dependent |
| GIF / APNG encoding | Supported | Supported | 16.4+ |
| Decode existing animation | Broadest support | Partial | Limited |
| HEIC read | Local WASM* | Local WASM* | Native 17+ / WASM* |
| AVIF write | Local WASM* | Local WASM* | Local WASM* |
| HEIC write | Server | Server | Server |
* Full self-hosted build: codecs load on demand and execute locally. Current Chrome or Edge is recommended for the broadest animation decode, WebGPU and WebCodecs support.
Text and vector artwork are already small. What pushes a file into the tens of megabytes is nearly always scanned pages, photos and screenshots. So ImgIng does not attempt a general re-layout — it spends everything on per-image decisions, which is exactly what ImgIng is best at.
| Dimension | ImgIng | Ghostscript in the browser · the common approach | Upload-based tools |
|---|---|---|---|
| Where it runs | On your device | On your device | Uploaded to a server |
| Runtime to download first | About 60 KB, loaded with the page | 10 MB+ of WebAssembly | None, but the file is uploaded |
| Resolution decision | Parses content streams and measures each image’s effective DPI | Also effective DPI (1.5× threshold) | Usually one setting for the whole file |
| Encoder choice | Races JPEG, indexed palette and grayscale lossless per image and keeps the smallest | One JPEG path for every colour image | Usually the same |
| Screenshots and charts | Indexed palette — usually smaller and sharper | Forced JPEG, prone to colour blotching | Forced JPEG |
| When nothing can be gained | Says so and keeps your original | Measured cases where the file grew | Depends on the implementation |
| Output verification | Re-reads the result; any page-count or page-box mismatch reverts to the original | None | None |
| Licensing | Entirely self-built, no AGPL obligations | AGPL-3.0 | — |
A 3000-pixel-wide image needs very different resolution filling a page than at business-card size. ImgIng parses each page’s content streams, tracks the graphics state matrix and derives the real placed size, then converts it to effective DPI. Only images clearly above the chosen preset are resampled — with Lanczos-3 rather than bicubic.
ImgIng reuses the same calibrated content classifier as its image compressor to tell screenshots, illustrations, icons, smooth photos and high-texture photos apart, then races JPEG, indexed palette with Flate, grayscale Flate and keeping the original. The workspace shows every image’s action, before/after size and measured DPI.
ImgIng walks the object graph from the document root and drops objects left by earlier edits, re-deflates uncompressed streams, strips layout-tool private data and page thumbnails, and repacks with object streams. Finally it re-reads its own output and reverts to your original if the page count or any page box does not match.
Honest limits: ImgIng does not subset fonts, so a text-only typeset PDF usually gains only 5%–20% from structural work and dedicated prepress tools will beat it there. ImgIng’s strong ground is scans, illustrated reports and exported decks — PDFs where images dominate. Encrypted PDFs are detected so you can unlock them first, and JPEG 2000, JBIG2 and CMYK print JPEGs are kept untouched because browsers cannot decode them reliably. Results depend on your file; we do not turn one sample into a universal claim — the workspace tells you image by image what it actually did.
ImgIng does not embed a PDF viewer or hide the source file inside a web page. Its self-built engine parses the document locally, reconstructs the page content as fixed-layout SVG, and exports one HTML containing only the text, graphics, images and permitted fonts needed to read those pages.
| Reconstruction area | HTML generated by ImgIng | PDF embed / viewer |
|---|---|---|
| Pages and text | Positions from page boxes, rotation, text matrices and spacing; text remains selectable | The PDF runtime still renders it |
| Special fonts | Checks OpenType embedding rights; rebuilds Type 3 glyphs from their vector outlines with a selection layer | Depends on the viewer's internal font handling |
| Graphics and colour | Rebuilds paths, clipping, inline and masked images, ICC / indexed colour, tiling patterns, axial / radial gradients and common soft masks | Interpreted live by the viewer |
| Result contents | One static HTML, no external dependencies, viewer controls or original PDF | Usually includes a toolbar, runtime or the source file |
| Safety and privacy | PDF stays on-device; PDF JavaScript is never executed | Depends on the viewer and deployment |
Parse the page tree, resources, content streams, graphics state, font maps, images, paths, links and form appearances instead of taking page screenshots.
Normal fonts become selectable text; Type 3 glyphs use their original CharProcs outlines; patterns, axial and radial gradients, images and common soft masks become inline SVG resources.
The source PDF exists only in the workspace for comparison. Non-equivalent special blending, mesh shading and interactive features are identified with exact page numbers; the final HTML contains neither the report nor an operation menu.
Honest boundary: this is faithful fixed-layout reconstruction, not automatic conversion into a responsive website. Isolated or knockout transparency groups, masks with special transfer functions or matte premultiplication, CMYK screen colour, cross-browser JPEG 2000, mesh shading, text clipping, multimedia and some interactive forms can still require review. The workspace identifies exact pages instead of silently claiming pixel-perfect equivalence.
A selected self-contained HTML appears immediately; authorize the complete folder for a page with local assets. Convert then shows five stages for assets, font and rendering readiness, Chromium output, automatic lossless minimization and file writing. @page stays exact; no-@page content becomes one custom full page. A4 or Letter applies only when explicitly selected.
Read a self-contained file directly or authorize the complete folder. Missing, remote or unstable resources appear in the conversion report instead of being silently ignored.
The source appears first. Conversion then settles fonts, images, charts and Canvas and chooses page geometry from @page, fixed-layout page metadata or the complete rendered webpage.
Chromium preserves CSS, web fonts, SVG / Canvas, selectable text, links and true page geometry. The generic PDF pass shares mathematically identical vector paths, compacts only glyph drawing proven to be permanently invisible, preserves every image pixel by default, and verifies page boxes, text and links by re-reading.
Processing boundary: import, folder access, asset mapping, preview and PDF optimization stay local. Only an explicit Convert action submits the script-free snapshot; temporary files are deleted and external DNS is blocked. Relative ES-module imports, Service Workers, cross-origin signed-in resources, nested iframes and DRM media still require review.
Each guide documents real settings, processing location, useful defaults and current limitations.
Decode iPhone photos locally and export broadly supported formats.
Read the conversion guide →PNG-8 · ALPHATune palette, dithering and multi-level transparency for graphics and UI assets.
Read the PNG guide →GIF · APNG · WEBPPreserve timing, edit frames and export several animation formats.
Read the animation guide →LOCAL AI · PRIVATERefine full alpha, set a colour or image background, arrange the subject and add a multilingual text watermark locally.
Read the AI guide →PP-OCRv6 · LOCAL WORKERThree model tiers load on demand and preserve text positions, confidence, source rows and table structure; copy and TXT use tabs between table columns.
Read the OCR guide and language limits →MULTI-PAGE · LOCALArrange images and multilingual text across independent pages, then save .imging v3 or export every page locally.
Read the online design guide →LAB-GRADE MATTING · BETARun quality-first temporal matting locally; the ultimate tier was best overall in current internal samples among comparable browser-side solutions tested.
Review the test scope and open →SELF-BUILT PDF ENGINEA self-built PDF engine measures each embedded image’s effective DPI from the content streams, then keeps the smallest of JPEG, indexed palette or lossless. PDFs are never uploaded and there is no multi-megabyte runtime to download.
See the method and open →CLEAN-PAGE HTMLRebuild selectable text, Type 3 glyphs, paths, masked images, ICC / indexed colour and axial / radial gradients locally; the clean single file opens with no viewer controls.
See the technical and quality boundaries →CHROMIUM PDFPreview HTML instantly, preserve true page geometry, then automatically optimize and directly save verified Chromium PDF bytes.
See conversion and asset boundaries →PRIVACY · VERIFIEDSeparate source-image upload, WASM delivery, AI model caching and clearly labelled server-only formats.
Read the data-flow guide →BENCHMARK · DISCLOSEDEvaluate size, quality, transparency and compatibility together.
Read the methodology →LOCAL VS CLOUDCompare privacy, animation, codecs and AI without inflated rankings.
Read the comparison guide →Visible answers match the page’s structured data and current product behaviour.
Common formats—including JPG, PNG, WebP, GIF and APNG—plus Image OCR and local AI cutout run in the browser. Source pixels and recognized text are not uploaded. Only clearly labelled specialist formats use a server.
Fast recognizes Simplified and Traditional Chinese, English, German, Spanish, Portuguese and French. Professional and Ultimate add Japanese. The current unified models do not recognize Korean. Layout-aware exports preserve visual rows and use tabs between detected table columns.
Device-side conversion, compression, animation and AI cutout are free, require no signup and add no watermark.
ImgIng reads PNG, JPEG, WebP, GIF, BMP, AVIF, HEIC and more, and exports still images plus GIF, APNG, animated WebP and animated AVIF.
The model must reach your browser before local inference can run. It is cached where possible and uses WebGPU first, falling back to WASM. Image pixels stay on-device.
Yes in the full build: libheif and libavif WASM load on demand, then run locally. HEIC writing remains a specialist server capability.
Squoosh excels at local still-image codecs; TinyPNG offers cloud compression and APIs. ImgIng integrates local-first stills, animation, HEIC / AVIF and local AI in one interface.
Choose a file, tune the result and download—without creating an account.
Drop or select one or many files, including HEIC, AVIF and animation.
Set format, quality, dimensions, palette, timing or AI model as needed.
Inspect size and detail, then save the real encoded result locally.
Device-side tools are free, require no signup and add no watermark.
↑ Choose an imageTell us what worked, which formats you need and what went wrong.
